RSG: Beating Subgradient Method without Smoothness and Strong Convexity
Authors: Tianbao Yang, Qihang Lin
JMLR 2018 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the effectiveness of the proposed algorithms on several machine learning tasks including regression, classification and matrix completion. Keywords: subgradient method, improved convergence, local error bound, machine learning. Section 8 presents some experimental results. |
| Researcher Affiliation | Academia | Tianbao Yang EMAIL Department of Computer Science The University of Iowa, Iowa City, IA 52242, USA Qihang Lin EMAIL Department of Management Sciences The University of Iowa, Iowa City, IA 52242, USA |
| Pseudocode | Yes | Algorithm 1 SG: bw T = SG(w1, η, T) ... Algorithm 2 RSG: w K = RSG(w0, K, t, α) ... Algorithm 3 RSG with restarting: R2SG |
| Open Source Code | No | The paper includes a license for the paper itself, but does not contain any explicit statement about releasing source code for the algorithms described, nor does it provide a link to a code repository. |
| Open Datasets | Yes | We conduct experiments on two data sets from libsvm website 6, namely housing (n = 506 and d = 13) and space-ga (n = 3107 and d = 6). ... We conduct the experiment on the dna data (n = 2000 and d = 180) from the libsvm website ... We consider a movie recommendation data set, namely Movie Lens 100k data 7, which contains 100, 000 ratings from m = 943 users on n = 1682 movies. |
| Dataset Splits | No | The paper mentions using several datasets for experiments but does not provide specific details on how these datasets were split into training, validation, and test sets, or specify a cross-validation methodology. It mentions using 'class 3 versus the rest' for dna data, which is a classification problem setup, not a dataset split specification. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory specifications) used to conduct the experiments. |
| Software Dependencies | No | The paper mentions implementing algorithms and conducting experiments but does not specify any software libraries, frameworks, or their version numbers that would be necessary for replication. |
| Experiment Setup | Yes | The value of α is set to 2 in all experiments. The initial step size of RSG is set to be proportional to ϵ0/2 with the same scaling parameter for different variants. ... The baseline SG is implemented with a decreasing step size proportional to 1/ τ, where τ is the iteration index. ... For R2SG, we start from t1 = 103 and restart it every 10 stages with t increased by a factor of 1.15. ... In our experiment, we set θ1,2,3,4 = (0, 3, 6, 9) and λ = 10-5 following (Yang et al., 2014). ... The strict lower bound fslb in Freund & Lu s algorithm is set to 0. |